Re: [v6ops] proposed TCP MSS text for rfc6204bis

Hi Warren,

Le 2012-05-21 à 16:49, Warren Kumari a écrit :

> [ Top posting / meta questions ]
> 
> So, both my memory and my google-foo are failing me…
> 
> Can anyone remember *why* the v6 in MTU is 1280 (and please don't say "Because the RFC says so!"…)

In the excellent French book in which I learned IPv6 in 2003 (livre.g6.asso.fr/index.php/Format_du_paquet_IPv6), here is what I learned:
"Choice of 1280 as minimum MTU is to permit IPv6 tunneling. 1500 octets is the generally admitted link MTU, that imposed by Ethernet." 

As this permits multiple tunneling layers without fragmentation on the most typical links, and limits to a moderate ratio waste of bandwidth on 1500-octet remote links, this is IMHO a valuable choice.

RD

> I have some niggling voice in the back of my head making 3GPP noises, but…
> 
> W
> 
> On May 20, 2012, at 10:04 AM, Mark Andrews wrote:
> 
>> 
>> In message <4FB89733.2080106@gmail.com>, Brian E Carpenter writes:
>>> On 2012-05-19 22:26, Randy Bush wrote:
>>>>> If you want to send packets of arbitrary size, in any environment
>>>>> where PMTUD is impossible or fails, won't you need to always include
>>>>> a fragmentation header in every packet greater than 1280?
>>>> 
>>>> see discussion of jumbo frames, commonly 4k or 9k, between consenting
>>>> adults on known links
>>> 
>>> Yes indeed, but that isn't the general case. Across the open Internet,
>>> I think we have the situation I described.
>>> 
>>> On 2012-05-19 21:16, Gert Doering wrote:
>>> 
>>>>>> UDP packets larger than 1280 bytes
>>>>>> Don't do that!
>>>> 
>>>> Tell that to the DNS people.  They seem to really like not-using-TCP.
>>> 
>>> Yes, but I understand that DNSSEC more or less dooms that plan anyway.
>>> 
>>> However, I thinks it's true that the only fail-safe solution is to
>>> include a frag header if you need to send UDP >1280.
>>> 
>>>  Brian
>> 
>> For DNS we just fragment at 1280 using IPV6_USE_MIN_MTU.   We were
>> thinking about this back in 1998 (draft-ietf-ipngwg-bsd-frag-00.txt)
>> which was rolled into the advanced socket api.  It took a few more
>> years than I would have liked to become RFC and for implementations
>> to be available.  EDNS was already being developed back then and
>> it was obvious that PMTUD wouldn't work for large nameservers even
>> if they got the ICMPv6 PTBs.
>> 
>> For DNS there is little to be gained by trying to send any bigger
>> packets.
>> 
>> YMMV for other UDP based protocols.
>> 
>> Mark
>> 
>>>   Brian
>>> _______________________________________________
>>> v6ops mailing list
>>> v6ops@ietf.org
>>> https://www.ietf.org/mailman/listinfo/v6ops
>> -- 
>> Mark Andrews, ISC
>> 1 Seymour St., Dundas Valley, NSW 2117, Australia
>> PHONE: +61 2 9871 4742                 INTERNET: marka@isc.org
>> _______________________________________________
>> v6ops mailing list
>> v6ops@ietf.org
>> https://www.ietf.org/mailman/listinfo/v6ops
>> 
> 
> -- 
> With Feudalism, it's your Count that votes.
> 
> 
> _______________________________________________
> v6ops mailing list
> v6ops@ietf.org
> https://www.ietf.org/mailman/listinfo/v6ops